The Gemini AI refuses to play chess against antique ATARI chess computers.

The Gemini AI refuses to play chess against the ATARI chess computer. And why does that system make that decision? There is a model in its algorithms that it uses to compare models and probabilities. And because two other AIs lost the match, that means it's more probable that Gemini loses the chess match than wins it. The other thing is that the AI is programmed to be polite and also serve commercial use, which means the AI translates that loss as bad for its own and its background company’s reputation. 

The reason why Gemini is considered polite is that the company wants users to like their product. This is the main problem with commercial AI development. The purpose of that thing is purely to make money for the companies. Not to serve the national interests or scientific work. This is why AI is sometimes misunderstood. They use things like mathematical statistics and other things to advance trust between it and users. And the other thing is that the AI must also support its users' willingness to select the AI service that benefits the company, behind the AI and LLM. 

The ATARI game consoles from the 1970s are not so easy to win, as we expect. If somebody played chess against those chess machines that used an interactive chessboard where the chess buttons had a digital ID. The player must also move the computer’s chess pieces. And the computer shows movements by using light pairs that point at the button. And then the system showed the point where the system wants to move the button when it notices something in those chessboards. The player must move the button as the chess computer wants. If the player didn’t follow the order, the system just repeated “that was not my move”. And refuses to continue the game. 

Or when we look at the discussion that the AI had about that match, the beginning was that the AI told how powerful it is, and how many moves it can calculate before, but then in real time, the AI refused. The AI could “think” that if it loses a chess game to some antique game console, that’s bad for business. 

The fact is that the ATARI is the RISC machine. Its only purpose is to play chess. There is a limited number of movements in chess. And that’s why chess is one thing that is used for AI development. The AI can make multiple models to make moves. But the AI must have knowledge of how to play chess. In the world of AI, that means the AI creates a new dataset for the action. And when AI expands its skills, that means it just makes or loads a new dataset for it. 

The AI will not think like we do. It creates datasets and combines data from different sources. Most of the hardware systems that run AI or language models are so-called neurocomputers. In a neural network, each computer can operate as part of the entirety or independently. The problem with every single neural network is that they need chess programs to play chess. In chess programs, every game or tactic that the system uses is a database or dataset. 

The next step is that the system must analyze each of the games stored in the chess program. And then that system must find the right game and find its counter game. Those games are tactics that the system must use. The problem is that the AI can find more games or datasets on the net if it has instructions for that thing. The AI doesn’t praise itself as we do. It simply tells things about the systems that run it, if it has the permission to give that answer. If that is not permitted, the AI can tell that it cannot answer. Or it can tell lies if there is a dataset that involves lies. The computer doesn’t even know if it is lying. The dataset involves all answers that the computer can give. 

https://www.freethink.com/artificial-intelligence/ai-datasets

https://futurism.com/google-ai-refuses-chess-atari

https://www.tomshardware.com/tech-industry/artificial-intelligence/google-gemini-crumbles-in-the-face-of-atari-chess-challenge-admits-it-would-struggle-immensely-against-1-19-mhz-machine-says-canceling-the-match-most-sensible-course-of-action